Multi-Objective Optimization of Reference Compound Lists for Rigorous Evaluation of Predictive Toxicity Models

This repository optimizes a set of compounds using a genetic algorithm. Furthermore, we curate a dataset of compounds used in pharmaceutical safety test validation studies by JaCVAM and compare them with the compounds derived from the genetic algorithm.

---

Publication

Not yet.

---

File Structure

The project is organized into the following main directories:

• data/: Contains data files.
• processed/: Processed data.
• raw/: Raw data.
• result/: The directory is structured to correspond with the figures in our paper.
• data_validation_test_and_HTS/: Contains validation test and HTS (High-Throughput Screening) data in an easy-to-use format.
• notebook/: Contains Jupyter Notebooks for various tasks.
• experiment/:
  • 0_dataset_unbalance.ipynb: Visualizes the imbalance in the number of positive data points and the bias in toxicity strength within the dataset.
  • 1_GA.ipynb: Runs the multi-objective genetic algorithm (GA) to optimize compound sets based on similarity scores and toxicity variation.
  • 2_UMAP.ipynb: Uses UMAP to visualize the distribution of compounds generated through a genetic algorithm.
  • 3_random_generation.ipynb: Compares the random population, GA results, and the validation test dataset by generating various plots and histograms.
  • 4_use_each_gen_for_eval.ipynb: Analyzes the performance of the GA by measuring the distance of the generated compound sets from the Pareto front.
  • 5_tox_prediction.ipynb: Compares toxicity prediction performance between the validation data and the GA-generated datasets.
• preprocess/:
  • 0_preprocess_ice.ipynb: Extracts LD50 or DART test data, including CAS-RN and toxicity information, from the ICE dataset.
  • 1_preprocess_toxcast.ipynb: Extracts Tox21 ER or AR test data, including CAS-RN and toxicity information, from the ToxCast dataset.
  • 2_preprocess_validation_test.ipynb: Prepares the main dataset for the genetic algorithm by removing compounds used in the validation test.
  • 3_preprocess_for_lookup.ipynb: Creates lookup tables containing canonical SMILES and physicochemical properties for compound pairs by utilizing PubChemPy and CAS-RN.
  • 4_preprocess_for_tox_prediction.ipynb: Prepares the datasets for toxicity prediction by splitting the data into training, evaluation, test, and GA-selected sets.
• src/: Python scripts for core functionalities.
  • ga.py: Genetic algorithm core components.
  • prep.py: Data preprocessing scripts.
  • util.py: Utility functions.
• requirements.txt: Lists the project dependencies.

---

Authors

• Yohei Ohto - Main contributor
• Tadahaya Mizuno - Correspondence

---

Contact

If you have any questions or comments, please feel free to create an issue on GitHub here, or email us:

• oy826c60[at]gmail.com
• tadahaya[at]gmail.com
• tadahaya[at]mol.f.u-tokyo.ac.jp